📈 Machine Learning for Time Series

From raw signals to reliable forecasts.

A hands-on course that takes you from the fundamentals of time series analysis to advanced forecasting and classification techniques. No slides, no fluff. Just real data, real models, and real evaluation.

---

Why this course?

Most ML courses treat time series as "just another dataset". This course treats it as what it is: a fundamentally different problem that requires specialised thinking.

You will learn how to decompose temporal signals, build robust forecasting models, evaluate them properly, and go beyond prediction into classification. Each notebook builds upon the previous one, creating a coherent path from first principles to production-ready techniques.

Learning Path

┌─────────────────────────────────────────────────────────────────────────────┐
│                    Machine Learning for Time Series                         │
├────────────────┬───────────────────┬───────────────────┬────────────────────┤
│   Foundations  │  Classical Models │    Evaluation     │     Advanced       │
│                │                   │                   │                    │
│  Introduction  │  Exponential      │  Metrics &        │  Advanced          │
│  Naive Models  │  Smoothing        │  Validation       │  Forecasting       │
│  Transforms    │  ARIMA Family     │  Model            │  Time Series       │
│  Decomposition │                   │  Comparison       │  Classification    │
└────────────────┴───────────────────┴───────────────────┴────────────────────┘

Quick Start

# Clone
git clone https://github.com/vortico/ml4ts-course.git && cd ml4ts-course

# Install (using uv, recommended)
uv sync

# Launch notebooks
jupyter notebook

Alternatively, install with pip:

pip install -e .
jupyter notebook

Course Contents

The notebooks are designed to be followed in sequence, each building upon concepts from previous chapters:

#	Topic	Description
01	Introduction to Time Series	Fundamental concepts: stationarity, autocorrelation, data structures, and visualisation techniques for temporal data
02	Naive Forecasting Models	Baseline models every forecaster must beat: naive, seasonal naive, drift, and average methods
03	Basic Transformations	Preprocessing for stationarity: differencing, Box-Cox transforms, and variance stabilisation
04	Decomposition Methods	Extracting trend, seasonality, and residuals using classical and STL decomposition
05	Exponential Smoothing	Simple (SES), Holt's linear, and Holt-Winters triple exponential smoothing with damped trends
06	Evaluating Forecasts	MAE, RMSE, MAPE, MASE; cross-validation strategies for time series; model comparison frameworks
07	ARIMA Family Models	AR, MA, ARMA, ARIMA, SARIMA: identification, estimation, diagnostics, and the Box-Jenkins methodology
08	Advanced Forecasting	Reduction to regression, recursive vs direct strategies, ensemble methods, and feature engineering
09	Time Series Classification	Distance-based, feature-based, and shapelet-based approaches to classifying temporal patterns

What's Inside

ml4ts-course/
├── notebooks/                  # Jupyter notebooks (the course)
│   ├── 01-introduction.ipynb
│   ├── 02-naive-forecasting-models.ipynb
│   ├── 03-basic-transformations.ipynb
│   ├── 04-decomposition-methods.ipynb
│   ├── 05-exponential-smoothing.ipynb
│   ├── 06-evaluating-forecasts.ipynb
│   ├── 07-arima-family.ipynb
│   ├── 08-forecasting-advanced-topics.ipynb
│   ├── 09-time-series-classification.ipynb
│   └── utils.py               # Shared helpers, custom transformers & loaders
├── data/                      # Real-world datasets (Parquet + CSV)
├── img/                       # Diagrams and figures
├── notes/                     # LaTeX course notes (PDF)
└── pyproject.toml             # Project config and dependencies

Datasets

Real-world datasets included for hands-on practice:

Dataset	File(s)	Domain	Description
Google Stock	google.parquet, GOOGL.csv	Finance	Historical OHLCV prices for Google/Alphabet
Australian Electricity	electricity_au.parquet, electricity_au_month.parquet	Energy	Half-hourly and monthly electricity demand from Australia
Energy Demand	energy_demand.parquet	Energy	Hourly energy demand patterns from Spain
Chemical Process	chemical_process.parquet	Industry	Sensor measurements from an industrial chemical process

Tech Stack

Layer	Tool	Purpose
Language	Python 3.9–3.13	Core language
Statistics	statsmodels	Classical statistical modelling (ARIMA, ETS, decomposition)
ML Toolkit	sktime	Unified interface for time series ML tasks
Auto ARIMA	pmdarima	Automatic ARIMA order selection
Forecasting	skforecast	Reduction-based forecasting with scikit-learn regressors
Visualisation	matplotlib, seaborn	Plotting and exploratory analysis
Data	pyarrow	Efficient columnar data storage and loading
Environment	Jupyter Notebook	Interactive learning and experimentation
Package Manager	uv	Fast, modern Python dependency management
Code Quality	ruff	Linting and formatting

Contributing

# 1. Clone and install
git clone https://github.com/vortico/ml4ts-course.git && cd ml4ts-course
uv sync

# 2. Make your changes in the notebooks

# 3. Submit a pull request

License

This project is maintained by Vortico for educational purposes.

---

Built with 🔥 by Vortico